yablak

15 ms·

151.
by yablak 5y ago
You mean "hacker news"?
152.
by yablak 5y ago
Looking for an escalation path w/in Geo.
153.
by yablak 5y ago
I like this code: - it's got a history/blame with associated bugs linked in the changes, including all the reasoning behind every line. - it can be versioned into rollouts and rolled back if it interacts with another system in a b
154.
by yablak 5y ago
https://archive.md/TaTd8
155.
by yablak 5y ago
I've been looking at this for the last couple of days. The ones I've seen that are relatively cheap include orange pi and nanopi. Nanopi seems to have slightly better adoption, but orange pi is not far behind, is cheaper on Aliba
156.
by yablak 5y ago
Wonder how many VCs are thinking it but avoid leaving a paper trail?
157.
by yablak 5y ago
https://en.wikipedia.org/wiki/Survivorship_bias
158.
by yablak 5y ago
This is why I hacker news.
159.
by yablak 5y ago
Google photos
160.
by yablak 6y ago
Didn't I read about this in The Three-Body Problem?
161.
Why is numerical analysis still a big issue? Hint: it's the speed(yablak.substack.com)
23 points by yablak 6y ago|0 comments
162.
by yablak 6y ago
Charging a one time fee of $1 per account on credit card feels like it would keep most of the trolls out and reduce moderation costs. I miss MetaFilter.
163.
by yablak 6y ago
You'd be surprised how many of the old, painstakingly tested fp algorithms, have been reimplemented in DL frameworks. I ported many algorithms from cephes to Eigen for use in TensorFlow. Some I was able to make simd. Then the XLA fo
164.
by yablak 6y ago
Reading between the lines, it sounds like what you should do is make it as easy as possible to pivot early in your research process... Lowering the temperature and increasing the activation energy over time as you find the black swan. Conc
165.
by yablak 6y ago
At least here in CA, if you are on the grid but don't have a battery then you will go black with everyone else even if you have solar. This is a safety measure for the technicians from the electrical company fixing the issue. Accordi
166.
by yablak 6y ago
TMJ can cause tinnitus; specifically there is a nerve connecting your jaw to the areas near your ear. Damage to this nerve, e.g., caused by years of unmitigated clenching/grinding at night or by an injury, can cause this. A neurologi
167.
by yablak 6y ago
Jax is pretty popular in the research community. Also supports automatic gradient calculations, and vectorized map. Targets the GPU via the XLA backend.
168.
by yablak 6y ago
Why was this downvoted? It's absolutely true and a big topic in AI fairness and ethics work.
169.
by yablak 6y ago
1000bulbs.com
170.
by yablak 6y ago
Now that absl::Cord has been opensourced, I'd expect a release of protobuf that takes advantage of it. Up to you to use the API though.
171.
by yablak 6y ago
Use double quotes around your search string, it gives the same behavior.
172.
Why your multiprocessing Pool is stuck (it’s full of sharks)(pythonspeed.com)
1 points by yablak 6y ago|0 comments
173.
by yablak 7y ago
You can use tensorflow's tf.data, I think.
174.
by yablak 7y ago
Yes, the standard and expected behavior is that logical_op(NaN, anything) == false [except inequality] because NaN is not ordered with, or equal to, anything, including another NaN. Otherwise you may have the strange behavior that two erron
175.
by yablak 8y ago
The truth is that the best architecture depends on the problem, and you get a lot out of "graduate student gradient descent", i.e., hyperparameter search and fiddling with the architecture fine grained details. You'll want t
176.
by yablak 8y ago
The shampoo bottle CPAP is described here in diagram form: https://www.semanticscholar.org/paper/CPAP%3A-a-guide-for-cl...
177.
by yablak 8y ago
One of the authors here. Working on TensorFlow has been quite a trip! Here's a followup paper submitted to Usenix: https://www.usenix.org/system/files/conference/osdi16/osdi16... And here's a
178.
by yablak 9y ago
Given Milner's background, I would think this is has been an "open secret" in Silicon Valley investment circles. Is that the case?
179.
by yablak 9y ago
(I'm author of the TF rnn api & tf.contrib.seq2seq) There's a lot of work being done on this specific part. If you have a standard RNN architecture you want to run, you can probably use the cudnn code in tf.contrib.cudnn to g
180.
by yablak 9y ago
Some context would have been nice. UPC is made for massively parallel systems (supercomputers, etc): http://upc.lbl.gov/

More ›